Fast convergence of MoM-based package-board extraction via incremental eigen-AGMRES method

Chatterjee, Gourav; Das, Arkaprovo; Gope, Dipanjan

doi:10.1109/epeps.2014.7103620

Cited by 1 publication

(1 citation statement)

References 7 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, these techniques are useful for acceleration where a large block of the matrix remains unchanged and is augmented by small incremental sub-matrices, as is the case for incremental design. In contrast, in this work, expanding on [22], selectively chosen Eigen-vectors retrieved from a neighboring frequency solution are applied to expedite the convergence of the next frequency solution, in an Augmented GMRES framework [23]. The key idea behind augmentation is to add the approximate Eigen-vectors to deflate the corresponding Eigen-values from the spectrum so that overall convergence can be improved.…”

Section: Introductionmentioning

confidence: 99%

Learning-Based Fast Iterative Convergence of 3-D MoM via Eigen-AGMRES Method

Chatterjee

Das

Gope

2015

IEEE Trans. Antennas Propagat.

Self Cite

View full text Add to dashboard Cite

3D full-wave electromagnetic simulation using Method of Moments (MoM) under the framework of fast solver algorithms like Fast Multipole Method (FMM) is often bottlenecked by the speed of convergence of the Krylov-subspace based iterative process. This is primarily because the Electric Field Integral Equation (EFIE) matrix, even with cutting-edge preconditioning techniques, often exhibits bad spectral properties arising from frequency or geometry-based ill-conditioning, which render iterative solvers slow to converge or stagnate occasionally. In this paper, a novel technique to expedite the convergence of MoM matrix solution at a specific frequency is proposed, by extracting and applying Eigen-vectors from a previously solved neighboring frequency in an Augmented Generalized Minimum Residual (AGMRES) iterative framework. This technique can be applied in unison with any preconditioner. Numerical results demonstrate up to 40% speed-up in convergence using the proposed Eigen-AGMRES method.

show abstract

Section: Introductionmentioning

confidence: 99%